[11] RFR(S): 8209670: CompilerThread releases code buffer in destructor is unsafe
David Holmes
david.holmes at oracle.com
Tue Aug 21 07:08:29 UTC 2018
Hi Tobias,
This looks good to me too.
Thanks,
David
> Thanks Vladimir! Could I please get another review for this?
>
> Best regards,
> Tobias
>
> On 20.08.2018 19:57, Vladimir Kozlov wrote:
>> Looks good to me.
>>
>> Thanks,
>> Vladimir
>>
>> On 8/20/18 4:06 AM, Tobias Hartmann wrote:
>>> Hi,
>>>
>>> please review the following patch:
>>> https://bugs.openjdk.java.net/browse/JDK-8209670
>>>
>>> Since JDK-8198756 [1], compiler threads can be added and removed at runtime. C1 compiler threads use
>>> a temporary code buffer that is allocated once the thread is created and needs to be freed if the
>>> thread is removed to avoid a memory leak (see JDK-8205499 [2]). Freeing is currently done in the
>>> thread destructor after the thread was removed from the thread list. This means that although we are
>>> holding the CodeCache_lock, freeing of the code blob might interfere with non-locking safepoint code
>>> that is concurrently walking the code cache.
>>>
>>> The fix is to free the temporary buffers at the two places where compiler threads are removed:
>>> http://cr.openjdk.java.net/~thartmann/8209670/webrev.00/
>>>
>>> This should be safe because it partially restores the code to pre-JDK-8205499. Testing on
>>> hs-tier1-3, jdk-tier1-3 and hs-precheckin-comp passed.
>>>
>>> An alternative would be to disable UseDynamicNumberOfCompilerThreads and do a proper fix in JDK 12:
>>> http://cr.openjdk.java.net/~thartmann/8209670/webrev.01/
>>>
>>> Thanks,
>>> Tobias
>>>
>>> [1] https://bugs.openjdk.java.net/browse/JDK-8198756
>>> [2] https://bugs.openjdk.java.net/browse/JDK-8205499
>>>
More information about the hotspot-compiler-dev
mailing list